JPH0619714A - Example base inference method - Google Patents

Abstract

PURPOSE:To provide an example base inference method by which the number of retrieval examples is reduced and retrieval time is shortened through the use of a model example which does not include a mistake. CONSTITUTION:The example base inference method is to input the specification (a) of a problem (input of specification 1) at the time of solving the problem, to retrieve an example (b) which is most similar to the specification (a) among the plural examples (b), (d),... which are previously stored in an example base 2 in a prescribed priority 3 (retrieval of example 4), to alter the example (b) on a difference between the specification (a) and the example (b) based on an alteration rule 5 (correction of example 6), to output a correction plan (c) (output of correction plan 7), to add the model example obtained by extracting a success example from a corrected example to a storage at the time of adding the corrected example (addition of example 8). The number of retrieval examples is reduced by using the model example which does not include the mistake and retrieval time can be shortened with the constitution.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a case-based reasoning method, and more particularly to a case-based reasoning method applicable to diagnosis, design and planning of various plants.

[0002]

2. Description of the Related Art Case-based reasoning is a well-known technique for efficiently solving a problem based on past successes or failures and making corrections according to the problem. FIG. 5 is an explanatory diagram showing a schematic configuration in an example of a conventional case-based reasoning method, and FIG. 6 is a flowchart showing a problem solving procedure by the conventional case-based reasoning method. As shown in FIG. 5, the conventional inference method is mainly composed of the steps shown in the following []. In solving a problem by this method, first, the specification a of the problem is input [specification input 1]. Next, the specification a is collated with a plurality of past cases b, d, ... Stored in the case base 2 in advance, and a case b having the most similar condition is searched in a predetermined priority order 3 [case Search 4]. Then, based on the retrieved past case b, the part different from the specification a is modified using the modification rule 5 unique to the problem [case correction 6]. The correction plan c obtained by such a procedure is a system solution and is output [correction plan output 7]. This solution is actually applied to confirm the effect, or evaluated by confirming the effect by a simulation or the like, and then, together with the result (success or failure), the correction plan c is again set as the past case b in the case base 2 Remember [Addition 8]. The procedure for problem solving by this method is shown in FIG. That is, the steps [Specification input 1], [Case search 4], [Case correction 6], [Correction output 7], [Case addition] in the order of steps S1, S2, ... 8] is executed and the problem is solved. In such a conventional case-based reasoning method, since past cases are used, (1) reasoning time is fast (2) knowledge acquisition is easy (3) there is a prospect of success or failure before application. There are advantages. Therefore, the conventional method has been widely applied to diagnosis, design, planning, etc. of various plants.

[0003]

The conventional case-based reasoning method as described above has the following problems because the quality and quantity of the past case b stored in the case base 2 influence the performance. It was (1) There is a high possibility that an actual case includes a mistake, but there are cases where the mistake is not recognized. (2) The specifications tend to be stricter year by year, so in the case of many years ago, the current situation may not be met. (3) A huge memory is required if the case is added every time, but the larger the number of processes, the slower the inference speed of the duplicate case. In order to solve the problems in the conventional technique, the present invention improves the case-based reasoning method and reduces the number of search cases by using a model case that does not include an error, thereby reducing the search time. The purpose is to provide an inference method.

[0004]

In order to achieve the above object, the present invention, when solving a problem, searches for a case most similar to the above problem from among a plurality of cases stored in advance, and searches for the above problem and the above case. In a case-based reasoning method in which the case is corrected and output based on a predetermined rule with respect to a part different from the above, and the corrected case is added to the memory, a model case in which a successful case is extracted from the corrected case is It is configured as a case-based reasoning method characterized by adding to the above memory.

[0005]

According to the present invention, when solving a problem, a case most similar to the above-mentioned problem is retrieved from a plurality of pre-stored cases, and a different rule between the above-mentioned problem and the above-mentioned case is searched based on a predetermined rule. When the corrected case is output and the corrected case is added to the storage, a model case obtained by extracting a successful case from the corrected case is added to the storage. By using the model case that does not include an error, the number of search cases can be reduced and the search time can be shortened.

[0006]

Embodiments of the present invention will be described below with reference to the accompanying drawings for the understanding of the present invention. The following embodiments are examples of embodying the present invention and are not intended to limit the technical scope of the present invention. FIG. 1 is an explanatory diagram showing a schematic configuration of a case-based reasoning method according to an embodiment of the present invention, FIG. 2 is a list showing actual order cases, FIG. 3 is a list showing model cases, and FIG. The input / output format in each process showing the design example by the case-based reasoning method of the present embodiment is shown. Further, the same symbols are used for the elements common to the explanatory view showing the schematic configuration of the conventional case-based reasoning method shown in FIG. As shown in FIG. 1, the case-based reasoning method according to this embodiment is mainly configured by the steps shown in the following []. When solving a problem by this method, first, the specification of the problem is input [specification input 1], and then this specification a is collated with a plurality of cases b, d, ... Priority 3
The case b having the most similar conditions is searched [case search 4], and the changed part 5 is changed based on the searched case b using the change rule 5 [correction of case 6]. The correction plan c thus obtained is output [correction plan output 7] and evaluated, and then the correction plan c is stored in the case base 2 [addition of case 8]. All of these processes are the same as in the conventional example, and the procedure for problem solving by this method is also the same as in the conventional example (see FIG. 6). However, in the present embodiment, the case b stored in the case base 2 is referred to the model case without directly using the past case, and in connection with this, the failure case is not stored as the model case and only the successful case is stored. It is different from the conventional example in that it is extracted and stored. However, actually, the failure case may not be excluded at all, but may be added as a model case after being corrected as necessary. An example in which the inference method of this embodiment is applied to the quality design of thick plates will be described below. Thick plates are steel plates with a thickness of 6 mm or more, and are used for structural members such as ships, bridges, and tanks. In addition, quality design of thick plate means the chemical composition, processing method, and inspection standard corresponding to the size, application, strength, and product processing method (welding, bending, ...) Designated in the design specifications. And so on. This quality design has always been automatically designed for continuous orders, but it is difficult to automate it for discontinued orders and special orders, and this design work affects all of the steel plate manufacturing process. The heat training staff was in charge of the design in order to exert the influence.

First, the process of creating a model case will be described below. Generally, an order for a quality design of a thick plate can be expressed by an input / output format as shown in FIG. Here, the application processing classification is a collection of processing methods for each similar application. For example, the processing method is determined for applications such as pipes, water pipes, and structural pipes like K00. There are many bending processes for this water supply pipe,
Also, if cracks occur in the bent part, it may cause water leakage, so it is necessary to sufficiently guarantee the quality compared with structural pipes. Therefore, a special design (strengthening of components related to toughness, strengthening of inspection standards) is added only when the application is a water pipe. FIG. 3 shows a model case created based on this actual order case. In the model case, input items that do not affect the output are deleted. Moreover, in the model case, the quality design No. determined according to the specific material standard name and the plate thickness. The numerical values of and components are generally defined (that is, modeled). For example, (rank up: 1) or (highest quality) in the model case corresponds to this. Here, (rank up: 1) means that the rank is raised by one step from the component value defined by the standard. These specific values are determined in detail by the modification rule 5. An example of designing using this model case
FIG. 4 shows the order of each step in FIG. That is,
The specification a input in [Specification input 1] in FIG. 1 is given in the format of (1) in FIG. Similarly, the case b searched in [Case search 4] is in the format of (2) in FIG. 4, and the case b ′ in the middle of change in [Correction of case 6] is in the format of (3) in FIG. The amendment plan c output by the amendment plan output 7] is given in the format of (4) in FIG. Then, the success case (or the corrected failure case) is extracted from the correction plan c, modeled and stored in the case base 2. Therefore, the format of the successful case (or corrected failed case) added to the case base 2 in [Add Case 8] is the same model case as (2) in FIG. However, when the same case as the model case added to the case base 2 is stored first, it is not stored redundantly and only a new case is stored. As described above, in the present embodiment, a model case that does not include an error is defined, this is used for case search to perform a basic design, and details are designed by a change rule. As a result, duplication of cases is eliminated and the number of search cases is reduced, so that the search time can be shortened. In addition, since model cases are used, it is not necessary to consider the processing of inconsistent cases when searching for cases, and since the errors are not included, the validity of the derived solution is guaranteed. As a result, the problem can be solved reliably and promptly.

[0008]

Since the case-based inference method according to the present invention is configured as described above, the duplication of cases is eliminated and the number of search cases is reduced, so that the search time can be shortened. Moreover, since model cases are used, it is not necessary to consider the processing of inconsistent cases when searching for cases, and
Since the error is not included, the correctness of the derived solution is guaranteed. As a result, the problem can be solved reliably and promptly.

[Brief description of drawings]

FIG. 1 is an explanatory diagram showing a schematic configuration of a case-based reasoning method according to an embodiment of the present invention.

FIG. 2 is a table showing actual order examples.

FIG. 3 is a list showing model cases.

FIG. 4 is an input / output format in each process showing a design example by the case-based reasoning method of the present embodiment.

FIG. 5 is an explanatory diagram showing a schematic configuration of an example of a conventional case-based reasoning method.

FIG. 6 is a flowchart showing a procedure for problem solving by a conventional case-based reasoning method.

[Explanation of symbols]

 1 ... Specification input 2 ... Case base 3 ... Priority 4 ... Case search 5 ... Change rule 6 ... Case correction 7 ... Correction output 8 ... Case addition

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Kazuyoshi Maeoka 1 Kanazawa-machi, Kakogawa City Kobe Steel Works, Ltd. Kakogawa Works

Claims (1)

[Claims] 1. When solving a problem, a case most similar to the above-mentioned problem is searched out of a plurality of cases stored in advance, and the case is corrected based on a predetermined rule with respect to a part different from the above-mentioned problem. In the case-based reasoning method of outputting the above, and adding the corrected case to the memory, the model-based reasoning method is characterized in that a model case in which a successful case is extracted from the corrected case is added to the memory.